[Mikola Lysenko (mikolalysenko)]

Fixes #1671

Problem

On 2026-10-22, merge-group run 39840895419 failed in yarn-classic 1.22.22. Two tests in e2e_redirect_yarn_classic_build panicked because corepack yarn@1.22.22 --version failed once, during a DNS timeout on the Depot host (lookup cache.depot.dev ... i/o timeout 90 s earlier). require_yarn_classic probed only once, and SOCKET_PATCH_YARN_E2E_REQUIRED=1 turns a failed probe into a panic. One blip:

• evicted Fix VEX product purl using raw PEP 440 version (#1589) #1653 from the queue, adding ~20 min before it merged
• forced CI: gem variant apply tests assert JSON codes (#1089) #1642's group, which had already passed, to rebuild (~12 min)
• cost ~94 Linux job-min (48.6 for the failed run + 45.7 for the discarded green run)

Change

crates/socket-patch-cli/tests/common/yarn_classic_vex.rs, require_yarn_classic:

• Retries the probe after 2 s and then 5 s, so 3 attempts in total, before it counts the release as unavailable.
• Prints each failed attempt's exit status and stderr. Before this change, stderr went to Stdio::null(). The SKIP/panic message now includes the last attempt's error.
• Does not retry when the corepack binary can't be spawned, because retrying can't help.
• A wrong release still panics on the first attempt.

I left out the script warm-up the issue also proposed. Each suite uses its own sandbox COREPACK_HOME, and the retry alone covers the observed failure. I didn't change the yarn-berry, pnpm or rush probes, to keep this PR to one topic.

Expected saving

At ~0.5 hits/day (n=1 so far), about 15 merge-queue min/day amortized on the queue's critical path and ~50 Linux job-min/day. A green run costs nothing extra: the retry only runs after a probe fails. PR feedback latency and macOS minutes don't change.

Measured result

• Locally, happy path: SOCKET_PATCH_YARN_E2E_REQUIRED=1 cargo test -p socket-patch-cli --test e2e_redirect_yarn_classic_build passed 40/40 in 22.4 s, the same as before the change.
• Locally, failure path: an unreachable registry (COREPACK_NPM_REGISTRY=http://127.0.0.1:9) with an uncached release (1.22.19). Each probing test logged 3 failed attempts with corepack's stderr, then still panicked under REQUIRED=1. That adds ~7 s per probe to a leg that is already failing.
• CI (this PR, head 28a5f04): all 54 checks passed. ci-ok and clippy are green. yarn-classic 1.22.22 took 1.87 min (20:41:30 → 20:43:22), within the baseline (p50 2.15 / p90 3.34 min, n=55), so the happy path is no slower. The saving comes from avoided merge-queue evictions, which show up only after merge. The profiler will check the eviction rate.

Where each test runs

All tests run where they ran before. No test moved, and no assertion was weakened. A release that really can't be fetched still fails the REQUIRED leg.

Risk

Low. The change touches only test-support code. Required checks ci-ok and clippy are unchanged.

https://claude.ai/code/session_016RS4Qh9aCHhzeMGKNok1oN

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

bugbot run

---

Generated by Claude Code

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 28a5f04. Configure here.}

[Mikola Lysenko (mikolalysenko)]

Burn-down agent: labeled Ready for review at 28a5f04.

• CI: 54/54 settled green on 28a5f04 (35 success, 19 skipped). Merges cleanly with main.
• Bugbot: reviewed 28a5f04, no findings.
• Scope: test-only change in crates/socket-patch-cli/tests/common/yarn_classic_vex.rs (retries the yarn-classic corepack probe so one DNS blip no longer evicts the merge-queue run, CI perf: yarn-classic 1.22.22 — one failed corepack yarn@1.22.22 fetch fails the leg and evicts a merge-queue entry (~30 MQ-min, ~94 job-min per hit) #1671). No CHANGELOG.md change.
• Slack: not announced in this run (no Slack send tool available); the next run will retry.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Final review brief

What it does. The yarn-classic e2e suites probe corepack yarn@1.22.22 --version before they run. This PR retries that probe up to 3 times (waits of 0 s, 2 s, then 5 s), so one DNS blip on a CI host no longer fails the REQUIRED=1 leg and evicts the merge-queue entry (#1671). Each failed attempt now prints corepack's stderr, which used to be discarded.

Risk: low. Only test-support code changes (1 file, +39/−23). The behavior that matters is kept: a wrong yarn version still panics on the first attempt, and a release that really can't be fetched still fails the REQUIRED leg, about 7 s later than before.

Look here

• yarn_classic_vex.rs:124:`` the retry loop and its backoff schedule
• yarn_classic_vex.rs:133:``` if the corepack binary can't be spawned, it stops right away with no retry
• yarn_classic_vex.rs:146:```` a version mismatch still panics without retrying
• yarn_classic_vex.rs:156:````` the final skip, whose message now includes the last error (skip still panics under REQUIRED, L103)

Verified

• Read the full diff and checked it against the issue and the description; they match.
• Checked all 14 call sites across the 3 yarn-classic test binaries. Only the fail path gets slower: up to ~7 s per probe, and the probes run in parallel threads.
• CHANGELOG.md is untouched, and nothing outside the one file changed.
• CI 54/54 settled on 28a5f04 (35 success, 19 skipped), with ci-ok and clippy green.
• Bugbot is clean on 28a5f04. There are 0 review threads.

Changes I made: none.

Open questions: none.

Auto-merge (squash) is armed, so approving sends it straight to the merge queue.

---

Generated by Claude Code